Image encryption and decryption method based on heterojunction resistive switching PUF

Abstract

The present application relates to a kind of image encryption and decryption methods based on heterojunction memristor PUF, belong to hardware security technical field, the method includes first entropy source extraction: extracting the high resistance state current of each unit in heterojunction memristor, second entropy source extraction: the same heterojunction memristor device is operated multiple times, reads its low resistance state current value and calculates its coefficient of variation, encryption structure dynamic binding: using average value and setting function dynamically determines global structure parameter in encryption algorithm, double binding encryption and decryption process: when encryption, with binary initial key as input, and according to global and local structure parameter, encryption operation is carried out, while recording the topological evolution index in encryption process;When decryption, the global and local topology is reconstructed by topological evolution index reverse, and decryption is completed after matching and obtaining binary initial key;It improves the ability of anti-brute force cracking, while guaranteeing the reversibility and reliability of encryption and decryption, with good scalability and application prospect.

Description

Technical Field

[0001] This invention belongs to the fields of novel memory technology and hardware security technology, specifically relating to an image encryption and decryption method based on heterojunction memristors (PUFs, Physics Unclonable Functions). Background Technology

[0002] Thanks to the rapid popularization of 5G communication and IoT technologies, the interconnection of lightweight terminals (such as smart wearable devices and sensors) is becoming increasingly frequent, leading to a sharp increase in the demand for the transmission and storage of sensitive data such as personal privacy and industrial control. These terminals typically have limited computing resources and strict power consumption constraints, which places higher demands on the lightweight nature of encryption schemes and hardware integration capabilities. Information security has become a core requirement for current terminal deployments.

[0003] Traditional encryption technologies are mostly built on classic algorithms such as AES and RSA, relying on pre-set keys or key management mechanisms in software. In lightweight terminal application scenarios, they are prone to problems such as key leakage, excessive power consumption, and insufficient structural flexibility. Based on existing variable-structure encryption and decryption algorithms, their structure control is mostly limited to pre-set templates in the software. The hardware selects the template through parameters, which cannot achieve deep binding between the software algorithm structure and the hardware. This results in weak resistance to structural modeling attacks and makes it difficult to meet the high security requirements of terminals.

[0004] Physically unclonable functions (PUFs), as a hardware-level security method, utilize hardware manufacturing differences to generate unique random responses, fundamentally solving the vulnerability of key pre-stored data. They offer advantages such as low power consumption and lightweight design, making them a core research direction for IoT terminal security. Among these, oxide-based memristors, due to their material flexibility and compatibility with CMOS processes, are widely used in constructing PUF modules. They provide a physical entropy source using the device's unique physical characteristics (resistance fluctuations, turn-on voltage differences, etc.), injecting hardware randomness into the encryption system. However, in traditional oxide memristor PUFs, the conductive filaments nucleate completely randomly at any location in the oxide layer, resulting in a trade-off between high randomness and low stability. Furthermore, existing encryption and decryption schemes based on memristor PUFs have significant drawbacks: on the one hand, the memristor PUF only serves as a key or random number generator and does not participate in the definition of the algorithm's internal structure. The algorithm structure is still dominated by software, making it impossible to achieve a unique binding between "hardware physical state and algorithm structure." On the other hand, even if attempts are made to adjust the algorithm structure using memristor parameters, it is only limited to "selecting a software-preset topology template" and does not achieve "dynamically generating a unique topology from the physical characteristics of the memristor." This results in the algorithm structure being able to be simulated by software, and the hardware binding effect is limited.

[0005] Therefore, how to optimize the PUF characteristics of oxide memristors and deeply integrate the physical characteristics (spatial distribution differences, temporal random fluctuations) of the PUF with the structural topology of encryption algorithms to achieve "unique binding between the algorithm structure and the physical state of the memristor array", while taking into account the low power consumption and high reversibility requirements of lightweight terminals, has become an urgent technical problem to be solved, which is of great significance for improving the encryption security of IoT terminals. Summary of the Invention

[0006] To address the problems of existing technologies, this invention provides an image encryption / decryption method based on a heterojunction memristor PUF. This scheme leverages the PUF characteristics of MXene-ZnO heterojunction memristors: MXene suspension is unevenly coated onto a ZnO thin film, and thermal annealing is performed to form an MXene-ZnO heterojunction film with a naturally random distribution of microstructure and conductive defect sites, creating distinct coated and uncoated regions within each device. The synergistic effect at the interface between MXene and ZnO enhances resistive switching stability while introducing controllable intrinsic random resistive switching characteristics. This allows the memristor to output a unique and unpredictable response signal under electrical excitation, thus constituting a physically unclonable function (PUF).

[0007] An image encryption / decryption method utilizing the PUF characteristics of MXene-ZnO heterojunction memristors: This image encryption / decryption method reads the physical characteristic parameters of the memristor array under different operating states, including the current difference between adjacent columns in the high-resistivity state and the current fluctuation of a single device in the low-resistivity state. These parameters are mapped to global and local structural control parameters in the encryption / decryption algorithm, respectively. The actual hardware characteristics drive the dynamic changes of the algorithm's internal structure. A binary key is generated by comparing the current difference between devices in the high-resistivity state and mapping the symbols. Data encryption is performed, and necessary mapping information and the final key stream generated during the encryption process are recorded. At the decryption end, the relative inverse operation is performed to accurately restore the original input data.

[0008] To achieve the above objectives, the technical solution of the present invention is as follows: an image encryption and decryption method based on a heterojunction memristor (PUF), wherein the method maps the two-layer physical entropy source of the same memristor to the key of the encryption algorithm and the reconfigurable topology, respectively. Specifically, the image encryption and decryption method includes:

[0009] First entropy source extraction: Extracting the high-resistivity current of each unit in the heterojunction memristor, using...

[0010] The current difference ΔI generated by differential comparison between adjacent devices is directly mapped to the sign of the binary initial key for encryption and decryption, and the average value of ΔI is calculated. ;

[0011] Second entropy source extraction: Perform multiple switching operations on the same heterojunction memristor device, read its low-resistivity current value and calculate its coefficient of variation σ. The switching operation of the MXene-ZnO heterojunction memristor array is a test cycle of no less than 100 SET and RESET cycle characteristic tests.

[0012] Dynamic binding of encrypted structures: utilizing the average value of ΔI And set the function The global structural parameters in the encryption algorithm are dynamically determined, including the round iteration round R, the number of blocks of input data in the absorption phase B, and the number of optional paths P; the local structural parameters in the encryption algorithm are dynamically determined using the setting function g(σ) of σ, including the number of diffusion module paths W, the flip ratio D, and the permutation strength S. The global and local structural parameters of the encryption algorithm are determined only by the entropy source ΔI and σ extracted from the current MXene-ZnO heterojunction memristor.

[0013] The dual-binding encryption / decryption process: During encryption, a binary initial key is used as input, and encryption operations are performed according to the aforementioned global and local structural parameters. Simultaneously, the topological evolution index during the encryption process is recorded. The dual-binding encryption / decryption process requires a binary initial key and a dynamically bound encryption structure; these two work together to complete the encryption operation. During decryption, the global and local topologies are reconstructed in reverse using the topological evolution index, and the binary initial key is obtained to complete decryption. This invention utilizes the uncontrollable spin-coating characteristics of MXene suspension in the spin-coating process, causing it to spontaneously form randomly distributed MXene films on ZnO films. The MXene coating areas differ between devices. Benefiting from edge effects and interface engineering, it improves resistive switching stability while introducing controllable intrinsic random resistive switching characteristics, overcoming the high randomness and low stability problems of traditional oxide memristors. Furthermore, the image encryption / decryption system based on this memristor array PUF fully utilizes the physical non-cloning and dual-entropy source characteristics of memristors to drive the internal system updates of the encryption / decryption method, improving resistance to brute-force attacks while ensuring the reversibility and reliability of encryption / decryption. It has good scalability and application prospects.

[0014] In the above scheme, ΔI and σ are taken from the high-resistivity state and low-resistivity state of the same device, respectively, and are used as the output of a single-device dual-entropy source.

[0015] In the first entropy source extraction, the current difference ΔI generated by the differential comparison between adjacent devices and its mapping to the key in the encryption / decryption method are as follows:

[0016]

[0017]

[0018]

[0019] Where ΔI is the current difference generated by the differential comparison between adjacent devices. The value of the high-resistivity current in the i-th row and j-th column is... This represents the high-resistivity current value in the i-th row and j+1-th column. Let be a binary mapping function for ΔI.

[0020] Where: mapping function g(σ) is a normalization function or a linear mapping function; the topological evolution index includes at least the sequence of values ​​of global structural parameters R, B, P and local structural parameters W, D, S during the encryption process.

[0021] The heterojunction memristor used is an MXene-ZnO heterojunction memristor prepared on a Pt-plated silicon wafer.

[0022] The MXene-ZnO heterojunction memristor has a structure consisting of a bottom electrode, a functional layer, and a top electrode from bottom to top; the functional layer is an MXene-ZnO heterojunction thin film.

[0023] The MXene-ZnO heterojunction includes a ZnO layer sputtered by magnetron sputtering and an MXene layer that is discontinuous and not densely coated on the ZnO layer. The two thin films together constitute the MXene-ZnO heterojunction structure.

[0024] In the MXene-ZnO heterojunction film, the MXene-covered area and the ZnO exposed area are randomly distributed alternately, and the boundary between them has an irregular random morphology. There are conductive defect sites at the heterojunction interface and in the ZnO exposed area of ​​the MXene-ZnO heterojunction film. The distribution density, size and position of the conductive defect sites are all randomly distributed.

[0025] The response signal output by the MXene-ZnO heterojunction memristor under electrical excitation is unique, constituting a hardware physically unclonable function (PUF).

[0026] In the MXene-ZnO heterojunction memristor, the MXene layer has a discontinuous coating structure, forming locally covered and locally uncovered areas. The MXene layer of different memristor devices has inherent differences in thickness, coverage position and coverage range. The inherent differences between MXene-ZnO heterojunction memristor devices have electrical differences, which constitute the hardware entropy source of the Physically Unclonable Function (PUF).

[0027] Thanks to the discontinuous coating structure of the MXene layer in the MXene-ZnO heterojunction memristor, local covered and uncovered regions are formed. Due to the unpredictable nature of random coating, thickness, and location, the MXene distribution of each device is naturally random and difficult to replicate. This setup increases the diversity of hardware entropy sources, and the local discontinuous structure enhances the variability between devices, effectively avoiding the problems of small device variability, limited entropy sources, and weak uniqueness and resistance to attacks caused by the continuous and uniform coating of traditional memristor PUFs.

[0028] Compared with the prior art, the advantages of the present invention are as follows:

[0029] 1. Randomness and Stability Optimization: This invention utilizes non-uniform MXene coating and the edge electric field concentration effect to actively confine filament nucleation to the MXene edge region, achieving spatial constraint on the nucleation location. At the same time, it preserves the inherent randomness of the edge geometry, ensuring the difference characteristics between different devices. These differences induce the memristor to generate random electrical responses under electrical excitation, forming the hardware entropy source of the physically unclonable function. This overcomes the problem in traditional oxide memristor PUFs where conductive filaments nucleate completely randomly at any position in the oxide layer, resulting in a situation where high randomness and low stability inevitably coexist and cannot be balanced.

[0030] 2. Structural variability: This invention uses the physical characteristic parameters of memristor arrays in two dimensions (space and time) to dynamically generate the global and local topological structures of encryption and decryption algorithms, enabling the overall evolution framework of the algorithm and the sub-block diffusion network to be reconstructed in real time according to the physical state of the hardware, thus breaking through the technical limitations of existing fixed algorithms that rely solely on the security of the key.

[0031] 3. Hardware Physical Binding Characteristics: This invention directly maps the physical characteristics of the memristor array PUF to the encryption and decryption algorithm topology generation rules, so that different memristor arrays correspond to unique global and local algorithm structures. At the same time, combined with the same source key generation mechanism, the algorithm structure and key are both driven by the hardware physical state, avoiding the problems of fixed algorithm structure and reliance on software randomness parameters in the prior art, and significantly enhancing the system's anti-cloning ability.

[0032] 4. Hierarchical Cooperative Control Mechanism: This invention drives global topology generation through spatial difference characteristics to regulate the overall evolution path of the algorithm, and drives local topology generation through temporal random fluctuation characteristics to enhance the perturbation effect of sub-blocks, forming a hierarchical cooperative control system that combines global structural control and local diffusion enhancement. This enables the encryption and decryption process to have dynamic adjustability in both the macro-evolutionary framework and the micro-perturbation mode, thereby improving the overall security performance of the system.

[0033] 5. Key structure homogeneity: This invention generates an initial key based on the physical differences of memristor arrays and synchronously drives global topology generation, so that the key and algorithm structure have the same source and are strongly bound to the same hardware physical state. This avoids the security inconsistency problem caused by the independent generation of key and algorithm structure in the prior art and improves the overall security and reliability of the system. Attached Figure Description

[0034] The accompanying drawings will be used to further illustrate the research scheme of this invention. The drawings below are only some examples of this invention and do not constitute any limitation on this invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0035] Figure 1 This diagram illustrates the internal structural changes of the SHAKE128 algorithm driven by the hardware PUF characteristics of the memristor.

[0036] Figure 2 The output shows the IV characteristic curve of a single memristor device at least 100 cycles with a current limit of 10mA, and the current distribution histogram of 200 (4 rows × 50 columns) memristor devices at a read voltage of 0.01V.

[0037] Figure 3 The low-resistivity cumulative resistance profile of a single device at 0.1V and the device at 10V. 4 The resistive state retention characteristic under s

[0038] Figure 4 A schematic diagram of differential generation of binary keys for a memristor array and a heatmap of the current distribution read from the array device in a high-resistivity state.

[0039] Figure 5 A graph illustrating the randomness and uniformity of differential binary key generation.

[0040] Figure 6 This paper presents the application of a structure-variable encryption / decryption method driven by the hardware PUF characteristics of memristors to image encryption / decryption. Detailed Implementation

[0041] To further illustrate the technical solution and advantages of the present invention, the present invention will be described below with reference to examples. It should be noted that the specific embodiments described herein are only used to explain the present invention and do not limit the present invention.

[0042] In response to the problems existing in the prior art, the embodiments of the present invention elaborate in detail how to integrate the characteristics of heterojunction memristor arrays (PUFs) into existing technologies and system architectures, intuitively demonstrating the great potential and significant benefits of this technology in practical applications, and providing a new path for upgrading security protection technologies in related fields.

[0043] Example 1: An image encryption / decryption method based on a heterojunction memristor (PUF), comprising two parts:

[0044] 1. Optimization of PUF characteristics of heterojunction memristors based on MXene non-uniformly coated ZnO.

[0045] 2. An image encryption and decryption method based on MXene-ZnO heterojunction memristor PUF.

[0046] Specifically, this invention will focus on the system design and mechanism implementation of the above two parts, forming a complete technical solution from hardware PUF construction to image encryption and decryption applications.

[0047] PUF characteristic optimization of heterojunction memristors formed by non-uniformly coated ZnO with MXene:

[0048] Using a spin-coating process, an MXene suspension is uniformly drop-coated onto a dense ZnO thin film. After thermal annealing, a randomly coated MXene film is formed. Electrodes are then deposited on a hard mask to fabricate a heterojunction memristor based on MXene randomly coated on a ZnO thin film. The microstructure and conductive defect sites exhibit a naturally random distribution. Through the interfacial synergy between MXene and ZnO, the heterojunction film enhances resistive switching stability while introducing controllable intrinsic random resistive switching characteristics. This allows the memristor to output a highly unique and unpredictable response signal under electrical excitation, thus constituting a physically unclonable function (PUF).

[0049] The MXene suspension is spin-coated onto a ZnO thin film, forming an MXene-ZnO heterojunction film that is discontinuous and non-dense. "Discontinuous" means that the MXene layer covers less than 100% of the ZnO film surface, with exposed ZnO surface areas existing between adjacent MXene regions; this results in locally covered and partially uncovered MXene areas. "Non-dense" means that gaps or pores exist within the MXene layer, and the MXene sheets are not tightly connected, forming a porous or open structure that allows external molecules to pass through. The thickness, coverage location, and coverage range of the MXene film on different devices inherently vary. These differences include variations in MXene film thickness, regionalized coverage locations, and varying coverage ranges. These inherent differences on different devices form the basis of the entropy source for MXene-ZnO heterojunction memristors, inducing random electrical responses in the memristors under electrical excitation, constituting the hardware entropy source of the physically unclonable function.

[0050] Specifically, on the non-uniformly coated ZnO thin film, each device has a portion covered by MXene and a portion not covered. Thanks to the edge effect and dielectric discontinuity, the electric field of the memristor array under electrical excitation is preferential to the MXene-covered portion and the uncovered portion, and its electric field strength is enhanced. The functional groups of MXene at the edge provide a pathway for the formation of conductive filaments and effectively control the formation of conductive filaments. However, the different shapes and sizes of MXene at different edge positions provide randomness for the formation and breakage of conductive filaments.

[0051] The edge structure formed by non-uniform MXene coating achieves electric field concentration through edge effect and dielectric discontinuity, and uses functional groups to provide filament pathways, thus realizing effective control of the filament formation position; at the same time, the inherent randomness of the edge geometry provides an intrinsic random source for filament formation and breakage, thereby achieving the unity of "controllability and randomness".

[0052] Image encryption and decryption method based on MXene-ZnO heterojunction memristor PUF.

[0053] Specifically, it includes the following steps:

[0054] Entropy source extraction specifically includes first entropy source extraction and second entropy source extraction:

[0055] A SET / RESET voltage signal is applied to the memristor to regulate its resistive switching state. After the device performance stabilizes, the current values ​​of adjacent devices in their high-resistivity states are read, differentially processed to form the current difference ΔI, and the average value is calculated. The coefficient of variation σ of the low-resistivity current of a single device.

[0056] The detailed steps are as follows: First, the device characteristics on the same chip are cyclically adjusted using SET / RESET electrical signals. Once the device performance meets the predetermined operating conditions, it is placed in a high-impedance state. Next, the corresponding current value is read. Using a differential comparison method, the current difference ΔI is generated by comparing the current values ​​of adjacent devices, and its average value is calculated. Then, a single device is continuously switched on and off for at least 100 cycles. The current value of the device in the low-resistance state is read and its coefficient of variation σ is calculated.

[0057] Dynamic binding of encrypted structures:

[0058] Based on the high-resistivity current extracted from the PUF characteristic parameters, the current difference ΔI value generated by the differential comparison between adjacent devices is mapped to the initialization input of the key generation module in the encryption / decryption method. The mapping rule is as follows:

[0059]

[0060]

[0061]

[0062] Where ΔI is the current difference generated by differential comparison between adjacent devices. The value of the high-resistivity current in the i-th row and j-th column is... This represents the high-resistivity current value in the i-th row and j+1-th column.

[0063] The current difference ΔI generated based on the high-resistivity current differential comparison results read from the adjacent columns essentially reflects the randomness difference between devices in the memristor hardware. It has a clear directionality and macroscopic stability, and is therefore used as a global structural control parameter for the variable addition method.

[0064] The aforementioned ΔI exhibits significant fluctuations, with a CV ≥ 0.2, to ensure sufficient differentiation in global control. Furthermore, the topological value range corresponds one-to-one with the physical state of the memristor array. Different arrays generate global structures with significant differences and unpredictability. The specific mapping rules are as follows:

[0065]

[0066] in ΔI is the average current difference, M is the number of rows in the memristor array, R is the number of iterations in the round, B is the number of blocks of input data in the absorption phase, and P is the number of optional paths. , For the minimum and maximum number of round trips, , The minimum and maximum number of blocks for the input data in the absorption phase. This represents the total number of possible paths. It can be a normalization function or a mapping of other functions.

[0067] The round-trip path R determines the iteration depth of state updates during encryption and decryption, and the range of the initial perturbation of the pixel propagating in the multi-round diffusion and permutation process is positively correlated with R, thereby enhancing the nonlinear relationship between plaintext and ciphertext.

[0068] The segment length B in the absorption phase affects the scale of data involved in encryption and decryption operations in each round. Its value is determined adaptively by the characteristics of the memristor PUF based on the scale of input data and security requirements. When the segment length is small, the local perturbation frequency is enhanced to meet the fine-grained security requirements. When the segment length is large, its strong cross-region diffusion capability achieves the overall security requirements of large-size data.

[0069] Unlike the number of rounds back R or the segment length B, the path selection parameter P does not directly change the perturbation intensity, but rather changes the evolution path of data or state during the encryption and decryption process. Based on the path selection parameter P, one or more state evolution paths are selected from multiple state evolution paths to participate in the current encryption and decryption process. This path selection mechanism complements parameters such as the number of rounds back and the segment length, enabling the algorithm to maintain functional consistency while possessing structural diversity, thereby improving its protection against structural modeling attacks and path prediction attacks.

[0070] Based on the above memristor array and its stability test of at least 100 cycles, the current of the device in the low-resistivity state is read and the coefficient of variation σ is calculated. It represents the transient noise of a single device. It is large in number and has significant microscopic randomness, which is suitable for controlling the local structural control parameters of the variable addition method.

[0071] The current read from a single device in the memristor array under low resistance conditions exhibits strong randomness and controllable amplitude fluctuations (CV≤0.2), ensuring sufficient microscopic rotation. During decryption, the same memristor array can repeatedly generate a consistent local topology. The algorithm's local structure is reversible and stable. The specific mapping rules are as follows:

[0072]

[0073] Where W is the number of diffusion module paths, D is the flipping ratio, and S is the displacement intensity. Number of available paths for the diffusion module , and For the minimum and maximum allowed flip ratio, and For the minimum and maximum replacement strength, It can be a normalization function or a mapping of other functions.

[0074] The number of diffusion module paths (W), the flipping ratio (D), and the permutation intensity (S) work together to determine the propagation mode and intensity of perturbations within a sub-block. Specifically, W changes the propagation path of the perturbation, D determines the perturbation coverage ratio, and S changes the spatial arrangement of the data, thus forming a highly irregular diffusion pattern at the local level.

[0075] As the values ​​of the aforementioned parameters change, the structural complexity and nonlinearity of the encryption and decryption process update accordingly, thereby affecting the difficulty for attackers to crack the encryption using brute force. By adaptively driving these parameter changes through the hardware PUF characteristics of the memristor, a dynamic correlation between the encryption / decryption structure and security strength is achieved.

[0076] Dual-binding encryption / decryption process:

[0077] The rules governing the mapping between the ΔI sequence and σ are applied to the corresponding modules in the encryption / decryption algorithm, driving the algorithm's internal state to dynamically change with the memristor hardware PUF, generating a variable-structure encryption / decryption process. Subsequently, the binary key generated using a differential method is used as the initial key input to perform encryption. Simultaneously, the inversion and flipping indices for each block in each round are recorded to form an encryption record. The decryption process then uses the indices and flipping operations recorded during encryption in reverse order, combined with the key and structural parameters generated from the mapping between ΔI and σ parameters measured from the same memristor array, to completely and reversibly recover the data content. Since the binary key and structural control parameters both originate from the physical characteristics of the same memristor array, any parameters deviating from the hardware output cannot correctly reproduce the encryption structure and key, achieving reversible decryption and a strong hardware correlation guarantee.

[0078] The current difference ΔI generated by the differential comparison between adjacent devices and the low-resistivity current variation coefficient σ are derived from the physical response of the memristor array under different operating states. ΔI is based on the current difference between adjacent memristor cells under the high-resistivity state, reflecting the spatial non-uniformity of the array devices, and its main feature is the randomness between devices. σ is based on the current fluctuation of a single memristor cell under the low-resistivity state during multiple readout processes, and is used to characterize the time stability of a single device.

[0079] Because their measurement states, statistical objects, and physical mechanisms differ, the same memristor array can ensure the controllability of local control parameters by satisfying that the coefficient of variation of ΔI is greater than or equal to a preset threshold. There is no conflict in parameter constraints between the two in their physical implementation. The ΔI sequence and σ are both derived from measured data of the same memristor array, forming the internal structure of the dual-entropy source-binding algorithm for the memristor PUF characteristics.

[0080] The structural control parameters must satisfy the statistical distribution conditions defined by the physical characteristics of the memristor array to ensure the consistency and reversibility of the internal state transition relationships of the encryption and decryption algorithms. If the structural control parameters are not derived from the ΔI and σ parameters of the corresponding memristor, but are generated by arbitrary external inputs, it cannot be guaranteed that the generated structural parameters will remain consistent during encryption and decryption. This may result in the inverse operation failing to cover the forward operation path, thus compromising the decryption reversibility.

[0081] The described key generation method generates a binary key as the data-level key input for the encryption / decryption algorithm. The structure control parameters are used to configure the structure of the encryption / decryption algorithm, and under these structural constraints, the binary key is used as the data-level key to perform encryption / decryption operations. Due to differences in the manufacturing processes of different memristor PUF devices, the generated physical random input data varies between different physical devices, making the key difficult to reproduce across different physical devices.

[0082] The memristor is an MXene-ZnO heterojunction memristor fabricated on a Pt-plated silicon substrate. The MXene insertion, acting as a functional layer, enables relatively stable current fluctuations in the low-resistivity state, meeting the requirements for local structural control parameters. The ZnO layer, acting as a resistive switching layer, utilizes the wide high-resistivity current range of the oxide memristor due to the random breakage of conductive filaments in the high-resistivity state, making it suitable for global structural control parameter conditions. The MXene-ZnO heterojunction memristor is a multi-row, multi-column array to ensure that the number of differential comparison current difference sequences between adjacent columns meets the entropy source requirement. In a preferred embodiment, a 4-row × 50-column array configuration is used.

[0083] The hardware feature parameters of this application not only act on the key input in the encryption process, but also directly act on the structure of the encryption algorithm, changing the encryption process from the source, making the hardware and software strongly coupled, and realizing a complete circuit of "hardware-algorithm structure-encryption and decryption".

[0084] This scheme features an infinite number of encryption and decryption structures, exhibiting stronger resistance to modeling. For the first time, it utilizes the spatial difference and temporal random fluctuation characteristics of memristor arrays to dynamically generate the global and local topologies of the encryption and decryption algorithms. This achieves a strong binding and reconstruction between the algorithm structure and the hardware physical state, thus overcoming the limitation of relying solely on key security within the traditional fixed algorithm framework.

[0085] Example 2: A specific embodiment of the present invention is provided, and an implementation scheme for this embodiment is provided: The flowchart of the variable encryption / decryption method based on the hardware PUF characteristics of memristors provided in this embodiment of the present invention is as follows. Figure 1 As shown, its content includes:

[0086] Step 1: Read Hardware PUF Characteristic Parameters: Apply a reset scan voltage of 0 to -1V to the MXene / ZnO heterojunction memristor array to keep all array cells in an initial high-resistivity state. Then apply a turn-on scan voltage of 0 to 2V to enable the devices to enter a low-resistivity state. Apply a 0.01V read voltage to obtain the current response in the low-resistivity state. Repeat this extraction multiple times (no less than 100 times) to obtain the low-resistivity current of a single device under cyclic characteristic testing and calculate the coefficient of variation σ. The current difference ΔI between adjacent columns in the high-resistivity state is the high-resistivity current read by applying a 0.01V voltage after different devices return to the high-resistivity state with a reset scan voltage of 0 to -1V, and obtained by differential comparison of adjacent columns. This is the statistical average value of ΔI. All the above readings of PUF characteristic parameters were obtained under the premise of stable device characteristics.

[0087] Step 2: Map ΔI and σ as global and local control variables in the encryption / decryption method, respectively. The specific mapping rules for ΔI are as follows:

[0088]

[0089] in, It is 2. It is 6. It is 256. It is 1024. It is 8.

[0090] The The normalization function has a range of 1 / 2π. The formula between them is:

[0091]

[0092] The specific mapping rule for σ is as follows:

[0093]

[0094] in, It is 8. It is 0.01. It is 0.1. =1, It is 5.

[0095] The This is a scaling function, with a range of... The formula between them is:

[0096]

[0097] Step 3: The encryption / decryption method used in this example corresponds to the SHAKE128 algorithm. The mapping rules are applied to the corresponding module in the encryption method. The binary key generated by the differential comparison mapping of adjacent columns is used as the initial key input in the encryption process to encrypt the input data. Subsequently, the decryption process is performed to recover the original data. This algorithm is presented as an embodiment using the memristor hardware PUF characteristics to drive the variable internal structure of the algorithm, and does not limit the encryption / decryption method or parameter settings used in this invention.

[0098] The hardware described in this embodiment is an MXene / ZnO heterojunction memristor fabricated on a Pt-plated silicon wafer.

[0099] The method for fabricating a memristor provided in this embodiment of the invention includes the following steps:

[0100] Step 1, Substrate Cleaning and Treatment: Using a 2-inch Pt-plated silicon wafer as the substrate, first rinse it with clean water and then dry it with a high-purity nitrogen gun; then place it in a beaker containing deionized water and clean it in an ultrasonic cleaner for 10 minutes to remove the surface oxide layer and impurities; after cleaning, wipe the surface of the silicon wafer with anhydrous ethanol and dry it again with a high-purity nitrogen gun to ensure the substrate is clean.

[0101] ZnO thin film preparation: A zinc oxide target with a purity of 99.99% was installed at the RF target position in the main cavity of the magnetron sputtering instrument, and a silver target was installed at the DC target position. The anode cover was closed. The treated Pt-plated silicon wafer was fixed to the sample tray with high-temperature resistant insulating tape and placed into the sample transfer chamber. The main cavity and sample transfer chamber were evacuated to 1.0 × 10⁻⁻⁻⁶. 5 Pa, send the sample tray into the main chamber; introduce a mixture of argon and oxygen (oxygen content 12%) at 36 sccm, set the sputtering power to 100W, turn on the zinc oxide target anode cover, sputter for 30 minutes to form a uniform ZnO film.

[0102] MXene film preparation and annealing: Prepare MXene suspension (Ti3C2 powder and DMSO solution are mixed at a concentration of 10 mg / mL and ultrasonically treated for 120 minutes to ensure uniform dispersion); fix the silicon wafer with ZnO film on the spin coater suction cup, drop-coat the MXene suspension, and set the spin coating program: rotate at 200 rpm for 30 seconds, then rotate at 2000 rpm for 60 seconds; after spin coating, place the Pt-plated silicon wafer on a rapid heating stage and anneal at 60°C for 120 seconds to evaporate the solvent and densify the MXene film.

[0103] Ag electrode preparation: The designed mask was attached to the annealed Pt-plated silicon wafer, and then placed back into the sample transfer chamber of the magnetron sputtering instrument. The vacuum was then evacuated to 1.0 × 10⁻⁻⁻⁶. 5 After Pa, the material is fed into the main cavity; 60 sccm of argon gas is introduced, the power is set to 80W, the silver target anode cover is turned on, and sputtering is performed for 5 minutes to form an Ag electrode on the surface of the MXene thin film, thus completing the MXene / ZnO heterojunction memristor.

[0104] like Figure 2 As shown, the performance of a single memristor was tested using a semiconductor analyzer. The results showed that the heterojunction memristor exhibited good C2C (cycle-to-cycle) consistency, making it suitable for subsequent experiments. Furthermore, through continuous reset and set processes, the heterojunction memristor array was able to stably reset back to a high configuration, and the corresponding current value was read. The current variation coefficient of 50.9% indicates that the current fluctuation range of the 200 memristor array is significant, making it suitable for constructing a ΔI sequence to map and control the variable parameters of the global structure. The above number is only an example; the size of the memristor array should meet the minimum entropy source requirement and can be arbitrarily configured according to actual needs.

[0105] Figure 3 The low-resistance current value read from a single device during cyclic characteristic testing (at least 100 cycles) shows a coefficient of variation of 15.8%, indicating controllable current fluctuations in the low-resistance state. The good randomness of σ satisfies the conditions for mapping control of local structure variable parameters. A reading voltage of 0.01V is applied, and the device is tested at 10... 4 The hardware maintains a stable resistance within a certain timeframe, and its non-volatility provides accurate stability for the subsequent coupling of data and algorithms.

[0106] The present invention provides a process for extracting a binary key based on the PUF characteristics of an MXene-ZnO heterojunction memristor array. The binary key is used as the initial data input for the aforementioned encryption / decryption algorithm. The specific steps include:

[0107] Step 1, Connect the Keithley 4200A-SCS Semiconductor Parameter Analyzer: Connect the 4200A analyzer to the memristor array on the Pt-plated silicon wafer;

[0108] Step 2, Memristor Cell Cyclic Stability Test: Repeat the SET / RESET cycle operation on all memristor array cells until the device performance is stable, and then put them back into the high impedance state;

[0109] Step 3: Read the current value of the memristor array unit in the high-resistivity state: Using the measurement function of the 4200A tester, read the current value of the 200 memristor array units on the Pt-plated silicon wafer in the high-resistivity state in sequence.

[0110] Step 4: Extract physical random numbers from differential current values: Compare the measured current values ​​corresponding to the high configuration of the 200 memristor array cells with adjacent columns according to the device cell position. Subtract the current value in the right column from the current value in the left column. If the difference is greater than 0, output 1 for that bit of the physical random number; otherwise, output 0. Repeat this process multiple times to generate a 100-bit physical random number. Figure 4 ); Figure 5 The Hamming weight and off-chip Hamming distance of physical random numbers generated by memristor heterojunction array (PUF) are shown. The average Hamming weight is 46% and the off-chip Hamming distance (inter-HD) is 48%, which shows the good randomness and uniformity of PUF.

[0111] Figure 6This invention relates to an application of a structure-variable encryption / decryption method driven by the characteristics of memristor hardware PUF in image encryption / decryption. In the encryption process, the ΔI and σ mapping extracted from the memristor hardware are used as global and local control parameters within the SHAKE128 algorithm, making the internal state updates of the encryption algorithm hardware-defined. Subsequently, the physical random number processed by high-resistivity current differential processing is used as the initial key input to the encryption algorithm to perform the XOR function, thereby achieving encryption and recording the index of each step of the encryption operation. In the decryption segment, based on the key finally generated in the encryption process, combined with the index records of each step in the encryption process, the inverse operation is performed to achieve data recovery.

[0112] The above description is merely a specific embodiment of the present invention, but its protection scope is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in this invention, based on the spirit and principles of the present invention, should be included within the protection scope of this invention.

Claims

1. An image encryption / decryption method based on a heterojunction memristor (PUF), characterized in that, The method maps the two-layer physical entropy source of the same memristor to the key of the encryption algorithm and the reconfigurable topology, respectively. Specific image encryption and decryption methods include: First entropy source extraction: Extracting the high-resistivity current of each unit in the heterojunction memristor, using... The current difference ΔI generated by differential comparison between adjacent devices is directly mapped to the sign of the binary initial key for encryption and decryption, and the average value of ΔI is calculated. ; Second entropy source extraction: Perform multiple switching operations on the same heterojunction memristor device, read its low-resistivity current value and calculate its coefficient of variation σ. The switching operation of the MXene-ZnO heterojunction memristor array is a test cycle of no less than 100 SET and RESET cycle characteristic tests. Dynamic binding of encrypted structures: utilizing the average value of ΔI And set the function The global structural parameters in the encryption algorithm are dynamically determined, including the round iteration round R, the number of blocks of input data in the absorption phase B, and the number of optional paths P; the local structural parameters in the encryption algorithm are dynamically determined using the setting function g(σ) of σ, including the number of diffusion module paths W, the flip ratio D, and the permutation strength S. The global and local structural parameters of the encryption algorithm are determined only by the entropy source ΔI and σ extracted from the current MXene-ZnO heterojunction memristor. The dual-binding encryption and decryption process: During encryption, a binary initial key is used as input, and the encryption operation is performed according to the global and local structure parameters mentioned above. At the same time, the topology evolution index during the encryption process is recorded. The dual-binding encryption and decryption process requires a binary initial key and a dynamically bound encryption structure. The two work together to complete the encryption operation. During decryption, the global and local topology are reconstructed in reverse through the topology evolution index, and the binary initial key is obtained after matching to complete the decryption.

2. The image encryption / decryption method based on a heterojunction memristor (PUF) according to claim 1, characterized in that, ΔI and σ are taken from the high-resistivity state and low-resistivity state of the same device, respectively, and are used as the output of a single-device dual-entropy source.

3. The image encryption / decryption method based on a heterojunction memristor (PUF) according to claim 1, characterized in that, In the first entropy source extraction, the current difference ΔI generated by the differential comparison between adjacent devices and its mapping to the key in the encryption / decryption method are as follows: Where ΔI is the current difference generated by the differential comparison between adjacent devices. The value of the high-resistivity current in the i-th row and j-th column is... This represents the high-resistivity current value in the i-th row and j+1-th column. Let be a binary mapping function for ΔI.

4. The image encryption / decryption method based on a heterojunction memristor (PUF) according to claim 1, characterized in that: Mapping function g(σ) is a normalization function or a linear mapping function; the topological evolution index includes at least the sequence of values ​​of global structural parameters R, B, P and local structural parameters W, D, S during the encryption process.

5. The image encryption / decryption method based on a heterojunction memristor (PUF) according to claim 1, characterized in that: The heterojunction memristor used is an MXene-ZnO heterojunction memristor fabricated on a Pt-plated silicon wafer.

6. The image encryption / decryption method based on a heterojunction memristor (PUF) according to claim 5, characterized in that, The MXene-ZnO heterojunction memristor has a structure consisting of a bottom electrode, a functional layer, and a top electrode from bottom to top; the functional layer is an MXene-ZnO heterojunction thin film. The MXene-ZnO heterojunction includes a ZnO layer sputtered by magnetron sputtering and an MXene layer that is discontinuous and not densely coated on the ZnO layer. The two thin films together constitute the MXene-ZnO heterojunction structure. In the MXene-ZnO heterojunction film, the MXene-covered area and the ZnO exposed area are randomly distributed alternately, and the boundary between them has an irregular random morphology. There are conductive defect sites at the heterojunction interface and in the ZnO exposed area of ​​the MXene-ZnO heterojunction film. The distribution density, size and position of the conductive defect sites are all randomly distributed. The response signal output by the MXene-ZnO heterojunction memristor under electrical excitation is unique, constituting a hardware physically unclonable function (PUF).

7. The image encryption / decryption method based on a heterojunction memristor (PUF) according to claim 6, characterized in that, In MXene-ZnO heterojunction memristors, the MXene layer has a discontinuous coating structure, forming locally covered and locally uncovered areas. There are inherent differences in the thickness, coverage location, and coverage range of the MXene layer between different memristor devices. Due to these inherent differences, MXene-ZnO heterojunction memristor devices have electrical differences, which constitute the hardware entropy source of the Physically Unclonable Function (PUF).

Images